Arxiv:2010.00040V2 [Astro-Ph.CO] 29 Jul 2021 Interferometry Detections of Gws (E.G., Abbott Et Al

Mon. Not. R. Astron. Soc. 000, 000–000 (0000) Printed 30 July 2021 (MN LATEX style ﬁle v2.2)

Bridging the gap: spectral distortions meet gravitational waves

Thomas Kite1?, Andrea Ravenni1†, Subodh P. Patil2‡ and Jens Chluba1§ 1Jodrell Bank Centre for Astrophysics, School of Physics and Astronomy, The University of Manchester, Manchester, M13 9PL, U.K. 2Instituut-Lorentz for Theoretical Physics, Leiden University, 2333 CA Leiden, The Netherlands

Accepted 2020 –. Received 2020 –

ABSTRACT Gravitational waves (GWs) have the potential to probe the entirety of cosmological history due to their nearly perfect decoupling from the thermal bath and any intervening matter after emission. In recent years, GW cosmology has evolved from merely being an exciting prospect to an actively pursued avenue for discovery, and the early results are very promising. As we highlight in this paper, spectral distortions (SDs) of the cosmic microwave background (CMB) uniquely probe GWs over six decades in frequency, bridging the gap between astrophysical high- and cosmological low-frequency measurements. This means SDs will not only comple- ment other GW observations, but will be the sole probe of physical processes at certain scales. To illustrate this point, we explore the constraining power of various proposed SD missions on a number of phenomenological scenarios: early-universe phase transitions (PTs), GW production via the dynamics of SU(2) and ultra-light U(1) axions, and cosmic string (CS) network collapse. We highlight how some regions of parameter space were already excluded with data from COBE/FIRAS, taken over two decades ago. To facilitate the implementation of SD constraints in arbitrary models we provide GW2SD. This tool calculates the window function, which easily maps a GW spectrum to a SD amplitude, thus opening another portal for GW cosmology with SDs, with wide reaching implications for particle physics phenomenology. Key words: cosmology: theory — gravitational waves — early Universe — inﬂation — cosmic background radiation.

1 INTRODUCTION cosmic string networks (Buchmuller et al. 2019). Given these exciting theoretical developments, it is interesting to ask which cos- Gravitational wave (GW) astronomy has become a reality. The now mological and astrophysical probes can help constrain these dif- routine detection of compact object mergers by the LIGO/Virgo ferent scenarios. In this paper, we show that CMB spectral distor- collaboration (Abbott et al. 2019) has made, for good reasons, the tions (SDs) can provide complementary information at frequencies study of GWs one of the most active and current topics in cos- f = 10−15–10−9 Hz unavailable to other probes. In this way, SDs mology and astrophysics. Ongoing and planned observations of the offer a bridge between scales probed by next-generation CMB sur- tensor perturbation power spectrum currently span some 21 orders veys (e.g., Ade et al. 2019; Hazumi et al. 2019; Delabrouille et al. of magnitudes of frequency: From cosmic microwave background 2019), and astrophysical GW observatories such as current (e.g. (CMB) upper limits on primordial B-modes (Ade et al. 2018; Perera et al. 2019) and future (e.g. Weltman et al. 2020) PTA mea- Aghanim et al. 2020) measurements at the lowest frequencies, to surements. arXiv:2010.00040v2 [astro-ph.CO] 29 Jul 2021 interferometry detections of GWs (e.g., Abbott et al. 2020b,a) and Pulsar Timing Array (PTA) measurements (e.g., Perera et al. 2019; How do CMB SDs constrain tensor perturbations at the scales Alam et al. 2020) at higher frequencies. In the next few years, a that they do? Spectral distortions are created by mechanisms that plethora of experiments will test different scales between these ex- lead to energy release into the photon-baryon fluid at redshifts tremes (e.g., Campeti et al. 2021, for overview). z . 2 × 106, when thermalization processes cease to be effi- Many physical processes can indeed lead to detectable tensor cient (Zeldovich & Sunyaev 1969; Sunyaev & Zeldovich 1970; perturbations (see Caprini & Figueroa 2018, for review). These in- Illarionov & Sunyaev 1975; Danese & de Zotti 1982; Burigana clude GWs from phase transitions (Caprini & Figueroa 2018; Nakai et al. 1991; Hu & Silk 1993; Chluba & Sunyaev 2012). Many et al. 2020), early universe gauge field production (Dimastrogio- sources of distortions exist within standard ΛCDM cosmology as vanni et al. 2017; Machado et al. 2019; Machado et al. 2020), and well as scenarios invoking new physics (see Chluba et al. 2019b, for broad overview), and innovative experimental concepts (Kogut et al. 2016, 2019; Chluba et al. 2019a) have now reached criti- ? E-mail: [email protected] † E-mail: [email protected] cal thresholds to significantly advance the long-standing distortion ‡ E-mail: [email protected] constraints from COBE/FIRAS (Mather et al. 1994; Fixsen et al. § E-mail: [email protected] 1996). A particular source of SDs is due to the dissipation of ten-

© 0000 RAS 2 Kite et al. sor modes while they travel almost unimpeded through the cosmic whereas during matter domination (MD), one finds plasma (Ota et al. 2014; Chluba et al. 2015a).  η2 k2 eq A(k) j (kη) + B(k)y (kη)2 if k > k How do tensor perturbations distort the CMB spectrum? In 0 2  η2 2 2 eq TGW(k, η) ≈  h i2 , general, perturbations in the photon fluid dissipate through electron  2 3 j2(kη) k kη if k < keq scattering and free-streaming effects. Dissipation of scalar pertur- 3 cos(2kη ) sin(2kη ) bations provides one of the guaranteed sources of SDs in the early A k − eq eq , ( ) = + 2 (3) Universe within the standard thermal history (e.g., Sunyaev & Zel- 2kηeq 2kηeq (kηeq) dovich 1970; Daly 1991; Hu et al. 1994; Chluba et al. 2012b,a). 1 cos(2kηeq) sin(2kηeq) B k − − − . ( ) = 1 + 2 2 Similarly, tensor modes lose a small fraction of their energy by con- (kηeq) (kηeq) 2kηeq tinuously sourcing perturbations in the photon fluid which then also Here, keq is the comoving wavenumber entering the horizon at the distort the CMB spectrum. In contrast to scalar modes, however, the time of matter-radiation equality ηeq, and j` and y` are the spherical dissipation is mainly mediated by free-streaming effects. As shown Bessel functions of first and second kind. For wavelengths much in detail by Chluba et al. (2015a), this leads to dissipation of per- smaller than those entering the horizon today (kη0 1) we can ex- turbations over a vast range of scales, extending far beyond those pand the GW transfer function derivatives at leading order in k. Ad- relevant to scalar perturbations. Thus, although the tensor dissipa- 0 2 ditionally, since we always observe quantities that involve (TGW) tion rate is suppressed relative to scalar dissipation (tensor modes integrated over some range of k, we can average over one period to are not significantly damped by interactions with the photons), this obtain (e.g., Caprini & Figueroa 2018) opens new avenues for model constraints from SDs. kη01 Building on Chluba et al. (2015a), we translate the relations ↓ hT 0 2i ≈ 2 4 between µ-distortions and primordial tensor perturbation into quan- GW(k, η) ηeq/2η , (4) tities commonly used for GW searches. This makes it easier to com- which is a smooth function of k valid during MD. Similarly, during pare SD limits to those from other probes. As examples we con- RD we can apply the same procedure to Eq. (2), and obtain sider several inflationary models which source GWs beyond vac- hT 0 2i ≈ 2 uum fluctuations, early-universe phase transitions (PTs) and cos- GW(k, η) 1/2η . (5) mic string (CS) networks, all of which demonstrate how SD mea- For later use we point out that during RD, where Eq. (5) is valid, surements are and will be important for excluding portions of their a ∝ η, while during MD relevant to Eq. (4) we have a ∝ η2. To- respective parameter spaces. Indeed we highlight that several of gether with Eq. (1), this means that the GW energy density at a −4 −2 the widely discussed models could have already constrained some given scale evolves as ΩGW ∝ a H ∝ const during RD and −4 −2 regions of their respective parameter spaces with SD limits from ΩGW ∝ a H ∝ (1 + z) in the MD era. COBE/FIRAS, taken over a quarter-century ago. Future spectrom- As pointed out in Watanabe & Komatsu (2006), the approxi- eter concepts like PIXIE (Kogut et al. 2016) and its enhanced ver- mations given above neglect some important details. One of these sions (e.g., PRISM Collaboration et al. 2014; Kogut et al. 2019) is the process of neutrino damping, which has its greatest effects could, through their increased sensitivity, significantly increase the on scales important to SD physics. The damping is effective dur- range of scales and parameter space covered. This could give CMB ing RD but only after neutrino decoupling (T . 2MeV), which spectral distortions an important role in this highly-synergistic taken together almost exactly coincides with the SD regime. This multi-messenger campaign, providing unique scientific opportuni- damping occurs since free streaming neutrinos correspond to a non- ties for the next generation of cosmologists and particle phenome- negligible fraction of the energy density of the Universe during RD nologists alike. and generate significant anisotropic stresses that result in the damping of tensor perturbations. The magnitude of the effect is a 35.6% decrease of the power available in GW (Weinberg 2004). To include 2 GWS IN THE EXPANDING UNIVERSE this effect the transfer function given in Dicus & Repko (2005) is A GW can be represented as a transverse traceless tensor pertur- used: bation of the metric’s spatial component, h , and the energy den- i j 0 1 X 0 0 h i ji TGW = an n jn(kη) − kη jn+1(kη) , (6) sity it carries is ρGW = hi jh /(32πG), where the prime denotes η conformal time derivatives1. If these GWs were produced primor- n even 2 −2 dially , we can define the GW fractional energy density per decade with the coefficients a0 = 1, a2 = 0.243807, a4 = 5.28424 × 10 −3 of wavelengths as (e.g., Watanabe & Komatsu 2006) and a6 = 6.13545×10 . This is valid for the range of scales needed in the following section. 1 ∂ρ (k, η) P (k) k, η GW T T 0 k, η 2 , It is clear from Eqs. (1), (4) and (5) that the exact transfer ΩGW( ) = = 2 2 GW( ) (1) ρc(η) ∂ln k 12a (η)H (η) functions are important quantities for comparing the effects of the where ρc is the critical density, and in the second equality we fac- GW background in the early and late Universe. In this paper, we tored the primordial tensor power spectrum PT and the determinis- compare the SD sensitivity (↔ early Universe) to PTA and inter- tic GW transfer function TGW. ferometry (↔ late Universe). Even CMB temperature anisotropies, In Watanabe & Komatsu (2006), several analytical approxima- although sourced early on, mostly probe the Universe after the RD- tions of the GW transfer function were developed. During radiation MD transition. Because of this it is essential to get the exact dynam- domination (RD) we have ics of this transition right for any comparison to be meaningful. To study the evolution of the GW background in detail we numeri- 0 2 2 2 TGW(k, η) ≈ k j1(kη) , (2) cally solved for the wave evolution through the RD-MD transition, which gave results agreeing with Watanabe & Komatsu (2006) and Dicus & Repko (2005) in the appropriate limits, while allowing us 1 We adopt the normalization conventions of Watanabe & Komatsu (2006). to more carefully model the GW background for a realistic cosmol- 2 We consider the case of sub-horizon generation further on. ogy involving neutrino and dark energy densities.

For the Planck 2018 best-ﬁt cosmology (Aghanim et al. 2020), the exact solution is well approximated by3

DΩrel −1 −3/2 −2 k ΩGW/PT = 1 + α1κ + α2κ + α3κ , κ = , (7) 12 k∗ −1 where α1 = 5.74, α2 = −2.47, α3 = 14.48, k∗ = 1/551 Mpc , D = −5 0.642 is the neutrino damping factor and Ωrel = 9.19 × 10 is the combined density of radiation and neutrinos, treating the latter as massless. This solution diﬀers by ∼ 20% from Caprini & Figueroa (2018), with better matching arising when neglecting the neutrino energy density. For details see Kite et. al. (in preparation). Since gravitational wave upper limits are usually quoted as function of frequencies rather than wavelengths, we will use the relation k/Mpc−1 = 6.5 × 1014 f /Hz to change units.

3 µ-DISTORTIONS FROM TENSOR PERTURBATIONS

Much like scalar perturbations, tensor perturbations dissipate over Figure 1. A series of curves demonstrating the form of the k-space window zmax time, both transferring energy to neutrinos and (in smaller propor- function Wµ for various upper limits in redshift. For practical purposes, 8 ∞ tion) to photons. Primordial tensor perturbations entering the hori- zmax = 10 is equivalent to Wµ ≡ Wµ. The solid curve is the result for zon during or slightly before the µ-era (5 × 104 . z . 2 × 106), power injection before the µ era, while the other curves suffer from some when dissipating, generate µ-distortions of the CMB that will be reduced visibility at higher k. The faded line shows the results without neu- observable today (Ota et al. 2014; Chluba et al. 2015a). trino damping, leading to a ' 30% increase across the window function. The average value of µ-distortions today is related to the pri- This means one can use the simpler versions of the transfer func- mordial tensor power spectrum via a window function Wµ(k) Z tion, valid in RD, given in Eqs. (2) and (5). The time dependence — which before was included in the physics underlying the win- hµGWi(η0) = d ln k Wµ(k) PT (k) , (8) dow function — has to be made more explicit. Using redshift z to which already averages oscillations by integrating over a transfer better match Chluba et al. (2015a), Eq. (8) can be generalized to function’s evolution throughout the µ-era, achieving the usual fac- Z ∞ Z ∞ tor of 1/2 implicitly. We calculate Wµ(k) numerically according to hµGWi(z = 0) = d ln k dz Wµ(k, z) PT (k, z) , (9) Chluba et al. (2015a). The window function is shown in Fig. 1. 0 0 In comparison to the corresponding k-space window function of where we introduced the GW-µ-distortion window primitive scalar perturbations (e.g., Chluba et al. 2012a, 2015b), the dissipa- Wµ(k, z), which captures the physics behind the damping of GWs. tion efficiency of tensors is about five orders of magnitude smaller, Note that with PT (k, z) = PT (k) we recover Eq. (8) by defining R ∞ W dz = W . The explicit form of the window primitive is4. highlighting how weakly tensor modes couple to the photon fluid. 0 µ µ Offsetting this loss, we can see that tensor modes contribute 2 2 8H η ∗ to the generation of µ-distortions over a vast range of scales, with a 0 2 −Γγη Wµ = 1.4 × TGW(k, z) TΘ(k, z) e Jµ(z) , (10) power-law decay of contributions at k & 106 Mpc−1 (Fig. 1). This is 45˙τ in stark contrast to the dissipation of scalar perturbations, which are and for convenience we summarize here the quantities which are limited to scales k ' 50 − 10, 000 Mpc−1, with a strong exponential relevant to calculate the window function (for their derivation and decay of contributions from k & 10, 000 Mpc−1 (e.g., Chluba et al. further explanation we refer to Chluba et al. 2015a):τ ˙ is the time ∗ −Γγη 2012a). Scalar modes damp by photon diffusion, which virtually derivative of the Thomson optical depth. The terms TΘe contain erases all perturbations once the dissipation scale is crossed. For the physics of how the GW transfer function TGW couples to the tensors, the photon damping is minute and photon perturbations photon fluid. These terms can be reliably approximated as are continuously sourced by the driving tensor force, explaining 1 + 4.48ξ2 + 91ξ4 this significant difference (Chluba et al. 2015a). This makes SDs a T k, z ≈ T ξ ≈ , Θ( ) Θ( ) 2 3 4 (11a) potentially unique probe of GWs from early-universe physics. 1 + 4.64ξ + 90.2ξ + 100ξ + 55ξ ∗ e−Γγη ≈ 1 , (11b)

3.1 Time-dependent injection 0 with ξ = k/τ . The final term Jµ(η) is the energy branching ratio, Equation (8) determines the SD signal from primordial perturba- which gives the fraction of total energy injected into the photon tions that were created during inflation and only later enter the hori- fluid that contributes to the µ distortion. We use the simple analytic zon to dissipate their energy. Another possibility is to have pertur- approximation of the branching ratio (‘method B’ in Chluba 2016): bations created on sub-horizon scales at later times. This requires a  5/2  −(z/zth) × 4 generalisation of the window function formalism to account for the e for z > 5 10 Jµ(z) ≈  , (12) new time dependence. 0 otherwise An immediate difference for sub-horizon injection is that neu- with z = 1.98 × 106 denoting the redshift where thermalisation trino damping will not occur, as this only matters for GWs that th becomes inefficient (see also Hu & Silk 1993). cross the horizon between neutrino decoupling and the start of MD.

√ 3 4 This does not include a factor of 1/2 for oscillations We match the notation of Chluba et al. (2015a) with TGW ≡ 2ηTh

We employ one further approximation in assuming the injec- Below we will consider the upper limit on µ-distortions set by −5 tion happens instantaneously across all scales at a time η∗. There- COBE/FIRAS (µ < 9 × 10 95%CL) (Mather et al. 1994; Fixsen −8 fore, the tensor perturbations are uncorrelated (PT (k, η < η∗) = 0) et al. 1996), and the forecasted constraints for PIXIE (µ < 3×10 ) −9 up to η∗ when their power spectrum abruptly jumps to some value (Kogut et al. 2011), SuperPIXIE (µ < 7.7 × 10 ) (Kogut et al. that we will now determine. The spectrum is found at all times after 2019), Voyage 2050 (µ < 1.9 × 10−9) and 10×Voyage 2050 (µ < −10 η∗ by redshifting ΩGW from the present-day value 1.9 × 10 ) (Chluba et al. 2019b), all of which already account for −4  a (η) 2 the presence of foregrounds following Abitbol et al. (2017).  2 ΩGW(k, η0) η > η∗ H (η) Ω (k, η) =  E (η) , E2(η) ≡ , (13) GW  2 0 η < η∗ H0 3.2 Scalar contributions −4 2 where we used a /E ∝ ρ¯GW/ρc. Above we discussed separating µGW from µother, taking the latter We will only consider power injection in the RD era, hence to be the standard model expected value. A second discussion is the tensor power spectrum is then obtained using Eq. (1) together necessary, however, regarding the contribution that scalar pertur- with Eq. (2), and reads bations have to a µ signal. This is important, considering that en-

 12H2 ergetic early-universe phenomena have the potential to generate  0  2 2 2 ΩGW(k, η0) η > η∗ scalar perturbations as well as tensors, which will enhance the SD P (k, η) =  a k [ j1(kη)] (14) T  0 η < η∗ production. In the following section we will discuss the scalar contributions for models where it is possible to do so, but some state- Notice that if not for the fact that the tensor perturbations ap- ments apply in general: Chluba et al. (2015a) show that the corre- pear at η , the power spectrum would always be time-independent: ∗ sponding window functions for scalar perturbations peak around it is in fact the equivalent of the primordial one in the “standard” 105 higher than for tensors, but for a narrower range of scales scenario described in Chluba et al. (2015a). (k ' 50 − 10, 000 Mpc−1 or f ' 8 × 10−14 − 1.5 × 10−11 Hz, as Using that Eq. (14) is independent of time during RD and in- previously discussed). Thus, for tensor perturbations to dominate serting this into Eq. (9), we can remove the time dependence in the spectral distortion signal the scalar spectrum created must be the integrand, leaving only changes in the upper limit of integra- less than 1 part in 105 of the tensor spectrum, or must be injected tion. It is therefore sufficient to study a series of window functions 4 −1 R z on smaller scales than k ∼ 10 Mpc . Provided both the wider ten- Wzmax = max W dz for different upper limits in time. Examples of µ 0 µ sor window, and that some early processes will be almost invisible Wzmax are shown in Fig. 1. We can observe that even modes origi- µ to scalar probes, the machinery explained above for constraining nating from z 2 × 106 contribute to the generation of distortions. early tensor energy injection are still of interest and importance, The flat plateau of the window function at k ' 0.1−103 Mpc−1 is not 5 zmax despite the relatively low sensitivity. affected until zmax . 5 × 10 , and will rapidly approach Wµ ' 0 4 as zmax approaches 5 × 10 . 4 MINIMALLY PARAMETRIC CONSTRAINTS For numerical applications, it is convenient to pre-tabulate zmax the tensor window function Wµ across k and injection redshift In this section, we calculate the constraining power of spectro- zmax. Since the background cosmology is fixed to high precision scopic CMB measurements in a minimally parametric fashion. As (Planck Collaboration et al. 2018), this procedure avoids additional in Campeti et al. (2021), we parametrize the primordial tensor approximations.5 However, a few comments are in place: we can power spectrum using logarithmically spaced tophat functions cen- further improve the treatment of the transition between the µ and tered around some ln ki with ln ki+1 − ln ki = 1.2 ∀i. This allow us y-distortion eras, which here we modelled as a step-function [see an easy comparison with Fig. 8 of their paper: Eq. (12)]. Including the more gradual transition (e.g., see discus- X PT (k) = AiWi(k) , (15a) sion in Chluba 2016), enhances the contributions from the largest i scales (k . 10−2 Mpc−1), however, a more accurate treatment of  1 if ln k ∈ [ln k − 0.6, ln k + 0.6] transfer effects is also required and left to future work.  i i Wi(k) =  . (15b) With the procedure outlined in this section we can calculate 0 otherwise the tensor dissipation contribution to the present day value of µ- Therefore, for each i, we insert Eq. (15a) into Eq. (8), and calcu- distortions. However, other processes, such as dissipation of acous- late the maximum value of Ai that is compatible with the chosen tic modes and Compton cooling also source µ-distortions, hence- hµGWi(η0) upper limit. With that information we then use Eq. (1) to forth referred to as µ . Any non-detection of an enhanced level of other calculate the corresponding ΩGW constraint. SD would straightforwardly constrain models that generate a large In Fig. 2 we show the sensitivity curves for COBE/FIRAS, µGW, comparable to or greater than µother. However, things are more PIXIE, SuperPIXIE, Voyage 2050 and 10×Voyage 2050, which delicate when µGW becomes much smaller than the value of µother all include estimated penalties from foregrounds. For comparison, −8 expected in the standard cosmological model, µother ' 2 × 10 we also report the sensitivity curves from Campeti et al. (2021), (e.g., Chluba 2016). In this regime, any actual analysis would re- which recently compiled the results of many planned experiments quire a marginalization of other sources, that we do not take into (Hazumi et al. 2019; Smith & Caldwell 2019; Arca Sedda et al. account here. However, assuming standard slow-roll inflation, we 2021; Sesana et al. 2019; Kuroyanagi et al. 2015; Crowder & Cor- can in principle accurately predict the expected standard contribu- nish 2005; El-Neaj et al. 2020; Reitze et al. 2019; Hild et al. 2011; tion given the power spectrum parameters measured at large angu- Weltman et al. 2020). Moreover, we show the NANOGrav 12.5 year lar scales (Chluba et al. 2012b; Khatri & Sunyaev 2013; Chluba & observation (Arzoumanian et al. 2020), interpreted as GW stochas- Jeong 2014; Cabass et al. 2016; Chluba 2016). For simplicity, we tic background according to their 5 frequency power-law model shall thus assume perfect removal of other µ-contributions. (see Kuroyanagi et al. 2020, for more discussion on whether the signal can be inflationary). Since the extrapolation of a red spectra 5 zmax A simple interpolation routine to calculate Wµ is available here: would be favourable for a SD detection, we conservatively assume https://github.com/CMBSPEC/GW2SD.git a flat spectrum.

1 k [Mpc− ] 4 3 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 10− 10− 10− 10− 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 2 10− 3 10− 4 10− FIRAS (95%CL) bin size 5 10− 6 10− 7 10− 8 PIXIE 10− Super PIXIE SKA 9 10− Voyage2050 LISA GW 10 10 Voyage2050 Ω 10− NANOGrav

2 × 11 ET h 10− 12 µ 10− Ares 13 DO Conservative 10− 14 DO Optimal 10− 15 10− AEDGE 16 DECIGO 10− LiteBIRD 17 10− 18 10− 9 8 7 6 5 4 3 2 1 0 1 2 3 4 19 18 17 16 15 14 13 12 11 10 − − − − − − − − − 10 10 10 10 10 − − − − − − − − − − 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 f [Hz] Figure 2. Upper limits on the energy density of gravitational waves from measurements of µ-distortions for various experimental configurations (COBE/FIRAS, PIXIE, SuperPIXIE, Voyage 2050, 10×Voyage 2050). For ease of comparison, we also report the upper limits for various other CMB, PTA and direct detection experiments (taken from Campeti et al. 2021), and the NANOGrav 12.5 years 95% confidence interval assuming a flat spectrum.

While the existing constraint derived from the COBE/FIRAS where the amplitude of tensor and scalar perturbations AT and data is a few order of magnitude higher than other probes, next AS are related by the tensor to scalar ratio by r ≡ AT /AS , and generation satellites will start to bridge nicely the frequency gap nT = −r/8 (Lyth & Riotto 1999). Current constraints, mostly existing between CMB observation and direct GW detection. It is driven by Planck low-` temperature and BICEP2/Keck B-modes interesting to notice that the upper bound from SDs will cover a data, (Ade et al. 2018; Aghanim et al. 2020) set the upper limit −1 very broad range of frequencies (more than 5 decades in f ). As r0.002 < 0.06 (95%) at k = 0.002 Mpc . Upon noticing |nT | ≤ such, any signal that is not sharply peaked in frequency will gener- 0.0075 ≈ 0, one can approximate the result by integrating a flat ate a comparatively higher µ-distortion, tightening the constraints spectrum yielding hµi(r) ≈ 1.68 × 10−13 r which gives the correct on specific parametric models, as we will see in the next section. result to within ≤ 5% for all values not ruled out by Planck. This shows that for any allowed value of r the SD signal will be out of reach for even the most sensitive SD mission concepts. 5 CONSTRAINTS ON SPECIFIC MODELS In principle this contribution is present as a component of tensor spectrum in the other models considered in the following In this section, we consider concrete models that generate GWs sections. However, since the amount of SD it generates is any- over a wide range of scales. For each of the following models it way negligible, we will omit it in the following. Note that the is enough to insert their corresponding tensor power spectrum into Planck constraint on r will also be considered for other models. Eq. (8) or (9) to obtain the predicted µ-distortion, depending on Strictly speaking, the aforementioned constraint only apply to a whether the injection is primordial or happens after reheating. power-law tensor spectrum, a condition not necessarily met by Generally speaking, once accounting for the limits on r from the models we will consider in the following. To provide some Planck (Ade et al. 2018; Aghanim et al. 2020), we understand context to the SD constraints we will draw, we opt to employ an that appreciable SDs can only be created by models with substan- order-of-magnitude estimate of the Planck constraint, simply re- tially enhanced tensor power at small scales. To also avoid fu- quiring that any spectrum of tensor perturbations, P (k), must sat- ture constraints at small scales, models with localized features at T isfy P (k)/P sf (k) < 0.06. In principle, a proper analy- f = 10−15 − 10−9 Hz are most promising. In the context of PTs, for T S k=0.002 Mpc−1 example, this identifies low-scale dark or hidden sector transitions sis of the Planck and BICEP2/Keck data could be carried out to set at energies ' 10 MeV - 10 eV in the post-inflation era as a target. constraints on the models that will be discussed here. This, however, goes beyond the scope of the paper. LiteBIRD (Hazumi et al. 2019), providing low multipole BB information at much higher pre- 5.1 Single-field slow-roll inflation cision, will allow us to further improve the limits set by Planck on As a benchmark we consider the tensor perturbations generated by the same range of scales in the near future. single-field slow-roll inflation. This model predicts a very low, almost scale invariant tensor spectrum, and as such we cannot expect 5.2 Spectator SU(2) axions SD constraints to be competitive with either CMB measurements Many inflationary models require the dynamics of additional spec- or future direct detections at small scales. We however include the tator fields active during the inflationary period, itself driven by a model for completeness, and as a point of comparison. The tensor separate scalar field. Generally speaking, the dynamics of the spec- spectrum from this model is given by tator field generate tensor perturbations in addition to those pro- sf nT PT = AT (k/k0) , (16) duced by the vacuum fluctuations of the quasi de Sitter background.

In this section, following Campeti et al. (2021), we consider It follows that the contribution of the oscillating axion and an axion-SU(2) spectator field based on the “chromo-natural” in- the subsequently produced dark photons must be sufficiently sub- flation model (Adshead & Wyman 2012; Dimastrogiovanni et al. leading to the energy density in the radiation fluid. Their relative 2017). Here, the SU(2) gauge fields acquire an expectation value, contributions to the curvature perturbation in total comoving gauge the fluctuations around which include a tensor perturbation with will consequently be suppressed relative to the contribution from a bilinear coupling to the graviton. The dynamics of the spectator fluctuations in the radiation fluid already present before dark pho- SU(2) axion are such that gravitons of a particular helicity are am- ton production (originating from the vacuum fluctuations sourced plified via a transient tachyonic instability, resulting in a (circularly during inflation). Hence the contribution of scalar perturbations polarized) contribution to the tensor power spectrum. sourced by axion dynamics to the µ−distortion signal will be sub- The spectrum for this model is given by (see Thorne et al. leading to those generated by primordial perturbations from infla- 2018) tion, and can safely be neglected. " !# This model is of particular interest to us as it produces a nar- 1 k P SU(2)(k) = r P sf(k) exp − ln2 , (17) rower spectrum of GWs. Thus, to constrain its parameter space it T ∗ R 2σ2 k p is important to have probes that can cover all phenomenologically which relates to the spectrum of scalar perturbations relevant frequencies. The GWs produced can be parametrized as a

sf nS −1 spectrum of the form PR = AS (k/k0) . (18) U(1) ˜1.5 In order to constrain this model, we use the best-fit Planck param- 6.3ΩGW ( fAA) k/k Ω U(1)(k) = , (19a) GW 1.5 h i eters (Planck Collaboration et al. 2018) for Eq. (18). However, r∗, 1 + k/k˜ exp 12.9 k/k˜ − 1 kp and σ are related to the parameters of the gauge theory and are essentially free to vary here. We take as a reasonable set of values, with those given in Campeti et al. (2021) (their model AX3) (r , k , σ) = ∗ p k˜ = 1.3 × 1015 f /Hz Mpc−1. (19b) (50, 106 Mpc−1, 4.8), which would yield µ = 2.1 × 10−12. Entertain- AA ing the question as to which set of parameters would maximize the Here ΩGW( fAA) and fAA are a function of the free parameters of the µ distortion signal while satisfying both observational and model model. These parameters, as introduced in Machado et al. (2019); 4 −1 constraints we find (r∗, kp, σ) = (265, 2.85 × 10 Mpc , 4.02). Machado et al. (2020), are fφ, m, α and θ, relating to the fit param- However, even this best case scenario leaves no appreciable SD eters in Eq. (19a) via signal, yielding µ = 2.1 × 10−11. This result can be understood by 2/3 1/2 −4 αθ m considering the parabolic shape of the spectrum in logP-logk space, fAA ≈ 6 × 10 Hz , (20a) which due to model constraints cannot peak to sharply (e.g. see 66 10meV " #4 " #4/3 Eqs. (A8) and (A11) in Thorne et al. 2018). This means that a spec- f θ2 U(1) ≈ × −4 −1/3 φ trum which avoids the Planck constraints cannot simultaneously ΩGW ( fAA) 1.67 10 gρ,∗ , (20b) Mpl α peak too high in the SD regime. In contrast, the models considered in the following subsections have spectra resembling broken which have both been redshifted to their present-day values. The power laws, and can be much more effective in satisfying current first two free parameters (i.e., fφ and m) essentially dictate the constraints while simultaneously generating significant SDs. height and frequency of the peak in the power spectrum respectively. The second two parameters are limited to α ∼ 10 − 100 and 5.3 Ultra-light U(1) audible axions θ ∼ O(1), and do not significantly change the shape of the spectrum for the range of allowed values. These parameters are therefore de- The sensitivity of SD measurements to axion standard model exten- generate with the first two. We choose fiducial values of α = 60 sions have already been discussed in the literature (Mukherjee et al. and θ = 1, but the main results given here hold more generally. 2018). In this subsection however we consider a model proposed in The direct dependence of fpeak on m means that different types Machado et al. (2019); Machado et al. (2020), in which the axions of experiment will probe different mass scales. This is shown Fig. 3, specifically produce a strong GW signal. In this scenario (generic where vertical dotted lines distinguish where different detection in the context of string compactifications), one has the presence of methods are dominant. From here it can be seen that SD are sensi- one or more U(1) axions with mass m and decay constant fφ that tive to the ultralight limit of the U(1) audible axion model, a result couple to dark sector photons. At early times during radiation dom- which again holds for any valid combination of α and θ. ination, when the Hubble parameter H is greater than m, the axion Note that Planck extends the limits from COBE/FIRAS at low field is over-damped and is frozen. Once H . m, corresponding to p masses to smaller values of fφ. Future SDs measurements could the temperature T ≈ mMpl, the axion starts to oscillate around significantly improve the limits from Planck to higher masses, cov- the minimum of its potential, sourcing gauge field production of a ering a wider range of the parameter space of phenomenological particular helicity that goes on to generate GWs. Since these GWs interest. are only produced on sub-horizon scales after the axion starts os- We note in particular how SDs can constrain masses in a range cillating, the results of Sect. 3.1 are essential in finding the µ signal not accessible to other measurements (10−22−10−13 eV). Such ultra- accurately. light axions may be ubiquitous in particular string compactifica- The audible axion scenario features a qualitative difference tions (Arvanitaki et al. 2010), and moreover, could be a viable dark with models that secondarily generate gravitation waves via gauge matter candidate were they to form a condensate at late times (Hui field production during inflation, as the generation occurs during et al. 2017; Marsh 2016), further illustrating the utility of SDs for radiation domination, when H < m. The oscillating axion at the particle phenomenology. minimum of the potential must remain a sub-dominant contribution to the energy budget, otherwise we’d have a phase of intermediate matter domination.

with peak frequencies

1 " # " # 6 β T∗ g∗(T∗) χ0 = , (21d) H∗ 100GeV 100 ! 0.62 f . × −5 χ , BC = 1 65 10 Hz 2 0 (21e) 1.8 − 0.13w + 3w −5 −1 fSW = 1.9 × 10 Hz 3w χ0 , (21f) −5 −1 fMHD = 2.7 × 10 Hz 3w χ0 . (21g)

Here, the three principal model parameters are α, β and 3w, which fix the amount of latent heat released by the transition as a fraction of the total energy density, inverse time duration of the PT, and velocity of bubble walls respectively. Denoted with ∗ are quantities at the time of the PT, making another key parameter zPT. The first two parameters follow 0 ≤ α ≤ 1 and β/H∗ > 1. The velocity of sound waves has been set to unity, since bub- Figure 3. A contour plot showing the expected SD signal arising from dif- ble walls usually propagate close to the speed of light. Parameters ∈ ferent combinations of fφ and m in the U(1) model. Without loss of general- labelled κi [0, 1] give the weighted contribution from each mech- ity, fiducial values of α = 60 and θ = 1 were chosen. Contours showing the anism. For this work we have used κBC = 1, κMHD = κv and visibility of several proposed spectrometers are shown. Vertical dotted lines α κv ≈ √ , (22) indicate regions of the phase space where different probes are most sensitive 0.73 + 0.083 α + α (from left to right: SD, PTA and Interferometry). Dotted lines continuing the SD mission contours show the estimates ignoring late time injection. the last of which is valid for 3w ' 1. The expected SD limits on PTs given these considerations are shown in Fig. 4. Even for low-energy PTs (α = 0.1) a PIXIE-like mission would explore some of the parameter space not already excluded by Planck; however, it would only see rather long PT. In the more energetic cases (α ≥ 0.5), SD missions could realistically detect PT lasting small fractions of the 5.4 Phase transitions beyond the Standard Model age of the Universe, and occurring relatively late in cosmic history. Evidently, SDs provide a unique and complimentary window The post-inflationary epoch may have seen a variety of first or- into low scale phase transitions (corresponding to energy scales in der phase transitions (PTs) in theories that go beyond the stan- the range 10 Mev - 10 eV) that are not possible to probe with any dard model of particle physics. First order PTs are characterized other observation. An important caveat to our discussion of this sce- by the fact that latent energy is released, and phases of true vacuum nario is the potential for the generation of sub-horizon scalar per- nucleate within false vacuum domains, resulting in bubble colli- turbations during and after phase transitions. Sub-dominant contri- sions (BC) that generate a stochastic GW background. Moreover, butions arising from the scalar field dynamics have been calculated magneto-hydrodynamic (MHD) turbulence and sound waves (SW) in Cutting et al. (2018), however retaining the scalar contributions in the bulk plasma during and after the phase transition also source from sound waves and MHD turbulence generated after the transi- sub-horizon GWs at commensurate frequencies. If these processes tion will require further study, and remains an important open ques- take place during the µ-era or shortly before, they can potentially tion for the present analysis. Our limits can therefore be considered result in measurable SDs. Here we once again use the results of conservative. Sect. 3.1 to calculate the associated SDs. Referring to the review of Caprini & Figueroa (2018), we see that the spectra resulting from the three different mechanisms for 5.5 GUT cosmic string networks GW production from PTs are given by Another tell tale sign of physics beyond the Standard Model is the existence of topological defects. Excluding textures, the standard 2 1 ! 2 ! 3 model does not allow for any defects. However, larger gauge sym- 2 BC −5 H∗ κBCα 100 h ΩGW( f ) = 1.67 × 10 (21a) metries (ubiquitous in models that go beyond the Standard Model) β 1 + α g∗(T∗) ! could admit symmetry breaking patterns that generate topological 0.1133 3.8( f / f )2.8 × w BC , defects in the early Universe (see Kibble 1980, 1982) which could 2 3.8 0.42 + 3w 1 + 2.8( f / fBC) have persisted into cosmologically observable epochs. Although 1 ! ! 3 the simplest models of monopoles and domain walls are tightly H∗ κvα 100 h2ΩSW ( f ) = 2.65 × 10−6 (21b) constrained (see Sects. 13.5.3 and 14.3.3 in Vilenkin & Shellard GW β 1 + α g (T ) ∗ ∗ 1994), cosmic strings networks remain a theoretical possibility and !3 ! 7 f 7 2 can impart potentially observable GW signals (see Sect. 10.4 of × 3 , w 2 fSW 4 + 3( f / fSW) Vilenkin & Shellard 1994). ! 3 ! 1 As an example, we consider a model proposed by Buchmuller H κ α 2 100 3 h2ΩMHD( f ) = 3.35 × 10−4 ∗ MHD (21c) et al. (2019) which attempts leptogenesis within an SO(10) grand GW β 1 + α g (T ) ∗ ∗ unified theory via a U(1)B−L phase transition, where a local U(1) ( f / f )3 baryon minus lepton number symmetry is spontaneously broken. × 3 MHD w 11 , 3 1 + ( f / fMHD) (1 + 8π f /h∗)

Figure 4. A series of contour plots showing the expected SD signal arising from low scale ﬁrst order phase transitions. Dotted lines (visible on the left) give the 5 6 sensitivity with standard window function, Wµ(k), showing that late power injection leads to a decrease of less than an order of magnitude for 10 < zPT < 10 . −10 The limits from Planck are also shown for comparison. The temperature at the time of the PT can be found with TPT/MeV ≈ 2 × 10 zPT.

The result of the B − L transition will be a meta-stable CS network generated at the time of the transition, which over the course of the collapse generates a mostly ﬂat spectrum of GWs due to the decay of string loops. An approximate form of their spectrum is given in terms of the model parameters κ and Gµ as6

2 CS 2 plateau h 3/2 i h ΩGW = h ΩGW min ( f / f∗) , 1 , (23a) Gµ −1/2 f = 3 × 1014 Hz e−πκ/4 , (23b) ∗ 10−7 Gµ 1/2 h2Ωplateau = 8.04 Ω h2 . (23c) GW r Γ Buchmuller et al. (2019) give a value of Γ ≈ 50 for this particular 2 −5 model, and we use a value of Ωrh = 2.5 × 10 . In reality string network collapse would be a function of time, but to match the formalism outlined in Sect. 3.1 we conservatively assume the entire spectrum emerges at the final moment of collapse given by Buchmuller et al. (2019) Figure 5. A contour plot showing the expected µ signal from a CS network !1/2 2 !1/4 70 (Gµ) −πκ arising from a U(1)B−L phase transition at the GUT scale. The limit placed zcollapse = Γ e . (24) by COBE/FIRAS is shown in red, and similarly for Planck in orange. Dashed H0 2πG contours show the sensitivity of various proposed SD missions. Faint dotted 3/2 The spectrum grows ∝ f up to f∗, and is flat for higher fre- lines show the contours without using the zmax limited window function. quencies. Furthermore, f∗ only depends weakly on Gµ but varies significantly with κ. This means that once κ is large enough that the spectrum is flat across the entire window of visibility for a given Although the generation of scalar perturbations of CS net- experiment, the probe will only be sensitive to Gµ. With SD mis- works in the scaling regime is well understood, the situation is sions probing lower frequencies than astrophysical probes they will much less clear for the scalar perturbations generated from the de- be complementary in limiting the lower bounds of the κ parameter. cay of meta-stable networks. Moreover, it is unclear whether the The potential of SD missions for constraining this model is shown dominant decay channel will be via gravitational wave production in Fig. 5. or scalar perturbations, and the answer will certainly be very model Given that the GW spectra produced by CS network collapse dependent. This remains another open question as far as this study has a plateau at smaller scales, for any given sensitivity depicted in is concerned, and the absence of any such detailed calculations is it- Fig. 2, we see that any one of the probes depicted will be equally self perhaps due to the fact that it may have been unclear in the past good at detecting the GW background produced. It is also worth what observational consequences, if any, sub-horizon generation of noting that the type of spectrum considered here will hold more scalar perturbations generated by CS network collapse would have. generally for a wide range of CS models (see Figueroa et al. 2020). We hope the results of this paper will provide the necessary moti- vation for such calculations.

6 Not to be confused with the SD amplitude µ. The combination Gµ will always be in reference to the energy scale of the CS physics.

6 DISCUSSION AND CONCLUSIONS As we have discussed, both in general and for specific models, any SD constraints should include both the scalar and tensor Highly energetic events in the early Universe, either during in- perturbations arising from energetic events (e.g. see Tashiro et al. flation or subsequently during radiation domination, can inject 2013; Amin & Grin 2014, for SDs from scalar perturbations of CS power into the GW spectrum. This can include GWs from sources and PTs, respectively). This potential for combining sources is an- within the standard ΛCDM cosmology, or from models that invoke other advantage SD experiments have over GW-based experiments, physics beyond it. Detecting the GW spectrum is therefore key to since the latter are only sensitive to the direct tensor perturbations. further scrutinizing our current paradigm, as well as pushing our This advantage is not easily utilised for the models discussed in this knowledge of the early Universe to new and exciting areas. Future paper, however, since the necessary scalar spectra are generally ab- experiments will probe these stochastic backgrounds, each sensi- sent from the literature. Where the inclusion is possible, it is again tive to a range of frequencies/wavelengths dictated by the nature important to highlight that SDs from tensor perturbations cover a of the experiment. As we have highlighted here, a wide range of wider range of physical scales than SDs from scalar sources, thus GW frequencies ( f = 10−15–10−9 Hz) can only be probed by SD extending the reach of SDs to earlier epochs. In addition, some sce- observations. This large span of wavelengths compensates for the narios do not produce any significant scalar perturbations [e.g., the relatively low efficiency of generating SDs from GWs, thus making axion-SU(2) model], making it crucial to account for SDs caused them a potentially powerful probe of physics beyond the Standard by tensor perturbations. Overall, SDs uniquely probe the presence Models of both particle physics and cosmology. of small-scale perturbations in regimes that are not directly acces- This work aims to introduce SDs as a complimentary probe sible, thus highlighting the important role that future CMB spec- through which one can detect and constrain stochastic GW back- trometers could play in GW cosmology, and, by extension, beyond grounds. The fundamental element to link these two messengers is the Standard Model phenomenology. the k-space window function, which maps a given GW spectrum into a SD signal imprinted before last scattering [see Eqs. 8 and 9]. In order to study the injection of power on sub-horizon scales, the window function for primordial tensor perturbations has been gen- DATA AVAILABILITY eralised [see Eq. 9], leading to minor changes in some models (Fig. Window functions (e.g. Fig. 1) are available at https://github. 3) but large changes in others (Fig. 5). This is essentially related com/CMBSPEC/GW2SD.git. to the fact that GWs have less cosmic history to dissipate their energy to the photon-baryon plasma. A simple python tool is provided 7 at GW2SD and allows one to easily estimate SD limits on various ACKNOWLEDGMENTS models, given the tensor power spectrum, PT (k, z), that comes into existence at a single redshift z. This is certainly a good approxima- We would like to thank Ana Achucarro and Jose Juan Blanco-Pillado for tion for 1’st order phase transitions, and holds to a good approxi- discussions about scalar perturbations generated during cosmic string col- mation for scenarios that dynamically generate GWs over a short lapse. We also thank Paolo Campeti and Eiichiro Komatsu for providing the duration. Refinements to account for the exact time-dependence of data used in Fig. 2, together with helpful discussions regarding the mapping the process are left to future work. between PT and ΩGW. This work was supported by the ERC Consolidator To illustrate the utility of SDs for GW cosmology, a series of Grant CMBSPEC (No. 725456) as part of the European Union’s Horizon phenomenological models were discussed, and their resulting SD 2020 research and innovation program. TK was further supported by STFC signals studied: As expected, the tensor perturbations generated by grant ST/T506291/1. JC was also supported by the Royal Society as a Royal single-field slow-roll inflation are too weak to be measured with Society URF at the University of Manchester, UK. SDs (Ota et al. 2014; Chluba et al. 2015a). Spectator axion-SU(2) fields too, even in more favourable cases that we considered, will realistically be out of reach in the foreseeable future. The Audible REFERENCES axion model (Sect. 5.3) on the other hand, can have a large region Abbott B., et al., 2019, Phys. Rev. X, 9, 031040 of its parameter space constrained by SDs, particularly for a wide Abbott B. P., et al., 2020a, The Astrophysical Journal Letters, 892, L3 range of masses in the ultra-light regime (Fig. 3). Similarly, the Abbott R., et al., 2020b, The Astrophysical Journal Letters, 896, L44 GWs from low scale (10 eV - 10 MeV) dark sector phase transitions Abitbol M. H., Chluba J., Hill J. C., Johnson B. R., 2017, Monthly Notices in the early Universe will be visible with future SD missions if the of the Royal Astronomical Society, 471, 1126–1140 relative energy content of the participating field is sufficiently large, Ade P., et al., 2018, Phys. Rev. Lett., 121, 221301 and the duration sufficiently long (see Sect. 5.4 and Fig. 4). The Ade P., et al., 2019, JCAP, 02, 056 typically flat GW spectra produced by CS networks can be seen by Adshead P., Wyman M., 2012, Phys. Rev. Lett., 108, 261302 many instruments, but SDs will be complementary to other probes Aghanim N., et al., 2020, Astron. Astrophys., 641, A6 in being sensitive to string collapse especially in the µ-era. It is Alam M. F. et al., 2020, arXiv e-prints, arXiv:2005.06490 noteworthy that some of the aforementioned models were already Amin M. A., Grin D., 2014, Physical Review D, 90, 083529 Arca Sedda M. et al., 2021, arXiv e-prints, arXiv:2104.14583 constrained with COBE/FIRAS long before first limits from Planck Arvanitaki A., Dimopoulos S., Dubovsky S., Kaloper N., March-Russell existed. Future CMB spectrometers like SuperPIXIE (Kogut et al. J., 2010, Phys. Rev. D, 81, 123530 2019) could establish a new frontier in this respect. Arzoumanian Z. et al., 2020, The Astrophysical Journal Letters, 905, L34 Buchmuller W., Domcke V., Murayama H., Schmitz K., 2019, arXiv e- prints, arXiv:1912.03695 Burigana C., Danese L., de Zotti G., 1991, Astronomy & Astrophysics, 246, 49 Cabass G., Melchiorri A., Pajer E., 2016, Physical Review D, 93, 083515 Campeti P., Komatsu E., Poletti D., Baccigalupi C., 2021, Journal of Cos- 7 https://github.com/CMBSPEC/GW2SD.git mology and Astroparticle Physics, 2021, 012

Caprini C., Figueroa D. G., 2018, Class. Quant. Grav., 35, 163001 Smith T. L., Caldwell R., 2019, Phys. Rev. D, 100, 104055 Chluba J., 2016, Monthly Notices of the Royal Astronomical Society, 460, Sunyaev R. A., Zeldovich Y. B., 1970, Astrophysics and Space Science, 227 9, 368 Chluba J. et al., 2019a, arXiv e-prints, arXiv:1909.01593 Sunyaev R. A., Zeldovich Ya. B., 1970, Astrophys. Space Sci., 7, 20 Chluba J., Dai L., Grin D., Amin M. A., Kamionkowski M., 2015a, Tashiro H., Sabancilar E., Vachaspati T., 2013, Journal of Cosmology and Monthly Notices of the Royal Astronomical Society, 446, 2871 Astroparticle Physics, 8, 35 Chluba J., Erickcek A. L., Ben-Dayan I., 2012a, The Astrophysical Jour- Thorne B., Fujita T., Hazumi M., Katayama N., Komatsu E., Shiraishi M., nal, 758, 76 2018, Phys. Rev. D, 97, 043506 Chluba J., Hamann J., Patil S. P., 2015b, International Journal of Modern Vilenkin A., Shellard E. P. S., 1994, Cosmic strings and other topological Physics D, 24, 1530023 defects Chluba J., Jeong D., 2014, Monthly Notices of the Royal Astronomical Watanabe Y., Komatsu E., 2006, Phys. Rev. D, 73, 123515 Society, 438, 2065 Weinberg S., 2004, Physical Review D, 69, 023503 Chluba J., Khatri R., Sunyaev R. A., 2012b, Monthly Notices of the Royal Weltman A., et al., 2020, Publ. Astron. Soc. Austral., 37, e002 Astronomical Society, 425, 1129 Zeldovich Ya. B., Sunyaev R. A., 1969, Astrophys. Space Sci., 4, 301 Chluba J. et al., 2019b, BAAS, 51, 184 Chluba J., Sunyaev R. A., 2012, Monthly Notices of the Royal Astronom- ical Society, 419, 1294 Crowder J., Cornish N. J., 2005, Phys. Rev. D, 72, 083005 Cutting D., Hindmarsh M., Weir D. J., 2018, Phys. Rev. D, 97, 123513 Daly R. A., 1991, The Astrophysical Journal, 371, 14 Danese L., de Zotti G., 1982, Astronomy & Astrophysics, 107, 39 Delabrouille J. et al., 2019, arXiv e-prints, arXiv:1909.01591 Dicus D. A., Repko W. W., 2005, Physical Review D, 72, 088302 Dimastrogiovanni E., Fasiello M., Fujita T., 2017, JCAP, 01, 019 El-Neaj Y. A., et al., 2020, EPJ Quant. Technol., 7, 6 Figueroa D. G., Hindmarsh M., Lizarraga J., Urrestilla J., 2020, Physical Review D, 102, 103516 Fixsen D. J., Cheng E. S., Gales J. M., Mather J. C., Shafer R. A., Wright E. L., 1996, The Astrophysical Journal, 473, 576 Hazumi M., et al., 2019, J. Low Temp. Phys., 194, 443 Hild S., et al., 2011, Class. Quant. Grav., 28, 094013 Hu W., Scott D., Silk J., 1994, The Astrophysical Journal Letters, 430, L5 Hu W., Silk J., 1993, Physical Review D, 48, 485 Hui L., Ostriker J. P., Tremaine S., Witten E., 2017, Phys. Rev. D, 95, 043541 Illarionov A. F., Sunyaev R. A., 1975, Soviet Astronomy, 18, 413 Khatri R., Sunyaev R. A., 2013, Journal of Cosmology and Astroparticle Physics, 6, 26 Kibble T. W. B., 1980, Physics Reports, 67, 183 Kibble T. W. B., 1982, Acta Physica Polonica B, 13, 723 Kogut A., Abitbol M. H., Chluba J., Delabrouille J., Fixsen D., Hill J. C., Patil S. P., Rotti A., 2019, in BAAS, Vol. 51, p. 113 Kogut A., Chluba J., Fixsen D. J., Meyer S., Spergel D., 2016, in Proc.SPIE, Vol. 9904, SPIE Conference Series, p. 99040W Kogut A., Fixsen D., Chuss D., Dotson J., Dwek E., et al., 2011, JCAP, 1107, 025 Kuroyanagi S., Nakayama K., Yokoyama J., 2015, PTEP, 2015, 013E02 Kuroyanagi S., Takahashi T., Yokoyama S., 2020, arXiv e-prints, arXiv:2011.03323 Lyth D. H., Riotto A., 1999, Phys. Rept., 314, 1 Machado C. S., Ratzinger W., Schwaller P., Stefanek B. A., 2019, JHEP, 01, 053 Machado C. S., Ratzinger W., Schwaller P., Stefanek B. A., 2020, Physical Review D, 102, 075033 Marsh D. J. E., 2016, Phys. Rept., 643, 1 Mather J. C. et al., 1994, The Astrophysical Journal, 420, 439 Mukherjee S., Khatri R., Wandelt B. D., 2018, Journal of Cosmology and Astroparticle Physics, 2018, 045 Nakai Y., Suzuki M., Takahashi F., Yamada M., 2020, arXiv e-prints, arXiv:2009.09754 Ota A., Takahashi T., Tashiro H., Yamaguchi M., 2014, Journal of Cos- mology and Astroparticle Physics, 10, 29 Perera B., et al., 2019, Mon. Not. Roy. Astron. Soc., 490, 4666 Planck Collaboration et al., 2018, ArXiv:1807.06209 PRISM Collaboration et al., 2014, Journal of Cosmology and Astroparti- cle Physics, 2, 6 Reitze D., et al., 2019, Bull. Am. Astron. Soc., 51, 035 Sesana A. et al., 2019, arXiv e-prints, arXiv:1908.11391